from the no-diving-in-the-shallow-learning-end dept.
http://www.nextplatform.com/2016/08/08/deep-learning-chip-upstart-set-take-gpus-task/
Bringing a new chip to market is no simple or cheap task, but as a new wave of specialized processors for targeted workloads brings fresh startup tales to bear, we are reminded again how risky such a business can be.
Of course, with high risk comes potential for great reward, that is, if a company is producing a chip that far outpaces general purpose processors for workloads that are high enough in number to validate the cost of design and production. The stand-by figure there is usually stated at around $50 million, but that is assuming a chip requires validation, testing, and configuration rounds to prove its ready to be plugged into a diverse array of systems. Of course, if one chooses to create and manufacture a chip and make it available only via a cloud offering or as an appliance, the economics change—shaving off more than a few million.
These sentiments are echoed by Naveen Rao, CEO of Nervana Systems, a deep learning startup that has put its $28 million in funding to the TSMC 28 nanometer test with a chip expected in Q1 of 2017. With a cloud-based deep learning business to keep customers, including Monsanto, on the hook for deep learning workloads crunched via their on-site, TitanX GPU cluster stacked with their own "Neon" software libraries for accelerated deep learning training and inference, the company has been focused on the potential for dramatic speedups via their stripped-down tensor-based architecture in the forthcoming Nervana Engine processor.
UPDATE: Intel bought the company for around $350 million to $408 million.
Related Stories
Intel will release a machine learning chip intended to compete with Nvidia's Tesla chips and Google's TPUs (Tensor Processing Units):
When the AI boom came a-knocking, Intel wasn't around to answer the call. Now, the company is attempting to reassert its authority in the silicon business by unveiling a new family of chips designed especially for artificial intelligence: the Intel Nervana Neural Network Processor family, or NNP for short.
The NNP family is meant as a response to the needs of machine learning, and is destined for the data center, not your PC. Intel's CPUs may still be a stalwart of server stacks (by some estimates, it has a 96 percent market share in data centers), but the workloads of contemporary AI are much better served by the graphical processors or GPUs coming from firms like Nvidia and ARM. Consequently, demand for these companies' chips has skyrocketed. (Nvidia's revenue is up 56 percent year on year.) Google has got in on the action, designing its own silicon named the Tensor Processing Unit to power its cloud computing business, while new firms like the UK-based Graphcore are also rushing to fill the gap.
Also at Engadget, TechCrunch, and CNET.
Previously: Nervana Deep Learning Chip
(Score: 2) by LoRdTAW on Wednesday August 10 2016, @05:07PM
So all it takes to make a shit load of money is to rehash an old idea and wait for a tech giant to buy it up.
(Score: 2) by takyon on Wednesday August 10 2016, @05:27PM
Or, the chip would have been so transformative that Intel bought the company to keep it off the market (without sporting an Intel sticker).
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by LoRdTAW on Wednesday August 10 2016, @08:05PM
Could be. Though, it looks like it happened pretty fast for what I consider a small sum of money seeing as how Intel threw 16.7 billion at Altera. Most likely Intel sees a good investment and Mr. Rao is being given a prominent position at Intel.
(Score: 2) by takyon on Wednesday August 10 2016, @08:32PM
I agree. Also, The Next Platform is reporting: [nextplatform.com]
I assume "quite a bit higher" means substantially more than $408 million.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]